Alignment of strings, EGCS all versions, gcc 2.95.*, i386

[Etienne LORRAIN]

Hello,

>> A very usefull soft:
 
bash-2.02$ cat alstr.c
void fct (char *param)
  {
  strcpy (param, "A very long string: 31 chars !");
  }

>> let's optimise the size:

bash-2.02$ gcc -Os -S alstr.c
bash-2.02$ cat alstr.s
        .file   "alstr.c"
gcc2_compiled.:
___gnu_compiled_c:
.text
        .align 32
LC0:
        .ascii "A very long string: 31 chars !\0"
        .align 4
.globl _fct
        .def    _fct;   .scl    2;      .type   32;     .endef
_fct:
        pushl %ebp
        movl %esp,%ebp
        pushl %edi
        pushl %esi
        movl 8(%ebp),%edi
        movl $LC0,%esi
        leal -8(%ebp),%esp
        cld
        movl $7,%ecx
        rep
        movsl
        movsw
        movsb
        popl %esi
        popl %edi
        movl %ebp,%esp
        popl %ebp
        ret
bash-2.02$

>> my main concern is with this line:
        .align 32

  Wasting, lets say an average of 16 bytes, for each
 string bigger than 31 chars (and some software have
 *a lot of* them) for saving very few cycles is boring.
 Most of those strings are a parameter of "printf" anyway...
 There is no way (no switch) to remove such alignment.

  In fact, this is more or less related to the message:
http://egcs.cygnus.com/ml/gcc-bugs/1999-11/msg00294.html
  where there was no answer.

  the alignment is done by macros "CONSTANT_ALIGNMENT"
 and "DATA_ALIGNMENT" in gcc-2.95.1/gcc/varasm.c

 IHMO those two macros should be removed from the
 default configuration of GCC and a *sensible* default
 value for the aligment of structures/arrays should
 be calculated. Then, if the user wants to, he can
 still set a very big, or very small, alignment of types
 with "attribute((align()))". That would remove the need
 of "attribute((packed))" for variables, which is
 documented but does not work anyway.

  Just hoping to be read,
  Etienne.
___________________________________________________________
Do You Yahoo!?
Achetez, vendez! À votre prix! Sur http://encheres.yahoo.fr